Question: AG-UI semantics for in-flight user steering

AG-UI docs mention "Agent steering", but I could not find protocol guidance for this case:

A run is already active. The user sends an additional instruction intended for that active run.

Example:

```json
{
"threadId": "thread-1",
"runId": "run-1",
"messages": [
{ "role": "user", "content": "Focus more on security tradeoffs." }
],
"forwardedProps": {
"steering": true
}
}
```

Questions:

1. Should this be modeled as a `RunAgentInput` request?
2. Should it reuse the active `runId`, or use a new `runId` with `parentRunId`?
3. What should the response look like?
- an ACK event stream,
- HTTP 204,
- `RUN_STARTED` / `RUN_FINISHED`,
- or something else?

Is there any recommended AG-UI pattern for this scenario?